freebsd-dev/contrib/groff/libbib/index.cc
1996-09-07 16:18:32 +00:00

624 lines
16 KiB
C++

// -*- C++ -*-
/* Copyright (C) 1989, 1990, 1991, 1992 Free Software Foundation, Inc.
Written by James Clark (jjc@jclark.com)
This file is part of groff.
groff is free software; you can redistribute it and/or modify it under
the terms of the GNU General Public License as published by the Free
Software Foundation; either version 2, or (at your option) any later
version.
groff is distributed in the hope that it will be useful, but WITHOUT ANY
WARRANTY; without even the implied warranty of MERCHANTABILITY or
FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
for more details.
You should have received a copy of the GNU General Public License along
with groff; see the file COPYING. If not, write to the Free Software
Foundation, 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. */
#include <stdio.h>
#include <string.h>
#include <stdlib.h>
#include <errno.h>
#include "posix.h"
#include "lib.h"
#include "cset.h"
#include "cmap.h"
#include "errarg.h"
#include "error.h"
#include "refid.h"
#include "search.h"
#include "index.h"
#include "defs.h"
// Interface to mmap.
extern "C" {
void *mapread(int fd, int len);
int unmap(void *, int len);
}
#if 0
const
#endif
int minus_one = -1;
int verify_flag = 0;
struct word_list;
class index_search_item : public search_item {
search_item *out_of_date_files;
index_header header;
char *buffer;
void *map_addr;
int map_len;
tag *tags;
int *table;
int *lists;
char *pool;
char *key_buffer;
char *filename_buffer;
int filename_buflen;
char **common_words_table;
int common_words_table_size;
const char *ignore_fields;
time_t mtime;
const char *do_verify();
const int *search1(const char **pp, const char *end);
const int *search(const char *ptr, int length, int **temp_listp);
const char *munge_filename(const char *);
void read_common_words_file();
void add_out_of_date_file(int fd, const char *filename, int fid);
public:
index_search_item(const char *, int);
~index_search_item();
int load(int fd);
search_item_iterator *make_search_item_iterator(const char *);
int verify();
void check_files();
int next_filename_id() const;
friend class index_search_item_iterator;
};
class index_search_item_iterator : public search_item_iterator {
index_search_item *indx;
search_item_iterator *out_of_date_files_iter;
search_item *next_out_of_date_file;
const int *found_list;
int *temp_list;
char *buf;
int buflen;
linear_searcher searcher;
char *query;
int get_tag(int tagno, const linear_searcher &, const char **, int *,
reference_id *);
public:
index_search_item_iterator(index_search_item *, const char *);
~index_search_item_iterator();
int next(const linear_searcher &, const char **, int *, reference_id *);
};
index_search_item::index_search_item(const char *filename, int fid)
: search_item(filename, fid), out_of_date_files(0), key_buffer(0),
filename_buffer(0), filename_buflen(0), common_words_table(0),
map_addr(0), map_len(0), buffer(0)
{
}
index_search_item::~index_search_item()
{
if (buffer)
free(buffer);
if (map_addr) {
if (unmap(map_addr, map_len) < 0)
error("unmap: %1", strerror(errno));
}
while (out_of_date_files) {
search_item *tem = out_of_date_files;
out_of_date_files = out_of_date_files->next;
delete tem;
}
a_delete filename_buffer;
a_delete key_buffer;
if (common_words_table) {
for (int i = 0; i < common_words_table_size; i++)
a_delete common_words_table[i];
a_delete common_words_table;
}
}
class file_closer {
int *fdp;
public:
file_closer(int &fd) : fdp(&fd) { }
~file_closer() { close(*fdp); }
};
// Tell the compiler that a variable is intentionally unused.
inline void unused(void *) { }
int index_search_item::load(int fd)
{
file_closer fd_closer(fd); // close fd on return
unused(&fd_closer);
struct stat sb;
if (fstat(fd, &sb) < 0) {
error("can't fstat `%1': %2", name, strerror(errno));
return 0;
}
if (!S_ISREG(sb.st_mode)) {
error("`%1' is not a regular file", name);
return 0;
}
mtime = sb.st_mtime;
int size = int(sb.st_size);
char *addr;
map_addr = mapread(fd, size);
if (map_addr) {
addr = (char *)map_addr;
map_len = size;
}
else {
addr = buffer = (char *)malloc(size);
if (buffer == 0) {
error("can't allocate buffer for `%1'", name);
return 0;
}
char *ptr = buffer;
int bytes_to_read = size;
while (bytes_to_read > 0) {
int nread = read(fd, ptr, bytes_to_read);
if (nread == 0) {
error("unexpected EOF on `%1'", name);
return 0;
}
if (nread < 0) {
error("read error on `%1': %2", name, strerror(errno));
return 0;
}
bytes_to_read -= nread;
ptr += nread;
}
}
header = *(index_header *)addr;
if (header.magic != INDEX_MAGIC) {
error("`%1' is not an index file: wrong magic number", name);
return 0;
}
if (header.version != INDEX_VERSION) {
error("version number in `%1' is wrong: was %2, should be %3",
name, header.version, INDEX_VERSION);
return 0;
}
int sz = (header.tags_size * sizeof(tag)
+ header.lists_size * sizeof(int)
+ header.table_size * sizeof(int)
+ header.strings_size
+ sizeof(header));
if (sz != size) {
error("size of `%1' is wrong: was %2, should be %3",
name, size, sz);
return 0;
}
tags = (tag *)(addr + sizeof(header));
lists = (int *)(tags + header.tags_size);
table = (int *)(lists + header.lists_size);
pool = (char *)(table + header.table_size);
ignore_fields = strchr(strchr(pool, '\0') + 1, '\0') + 1;
key_buffer = new char[header.truncate];
read_common_words_file();
return 1;
}
const char *index_search_item::do_verify()
{
if (tags == 0)
return "not loaded";
if (lists[header.lists_size - 1] >= 0)
return "last list element not negative";
int i;
for (i = 0; i < header.table_size; i++) {
int li = table[i];
if (li >= header.lists_size)
return "bad list index";
if (li >= 0) {
for (int *ptr = lists + li; *ptr >= 0; ptr++) {
if (*ptr >= header.tags_size)
return "bad tag index";
if (*ptr >= ptr[1] && ptr[1] >= 0)
return "list not ordered";
}
}
}
for (i = 0; i < header.tags_size; i++) {
if (tags[i].filename_index >= header.strings_size)
return "bad index in tags";
if (tags[i].length < 0)
return "bad length in tags";
if (tags[i].start < 0)
return "bad start in tags";
}
if (pool[header.strings_size - 1] != '\0')
return "last character in pool not nul";
return 0;
}
int index_search_item::verify()
{
const char *reason = do_verify();
if (!reason)
return 1;
error("`%1' is bad: %2", name, reason);
return 0;
}
int index_search_item::next_filename_id() const
{
return filename_id + header.strings_size + 1;
}
search_item_iterator *index_search_item::make_search_item_iterator(
const char *query)
{
return new index_search_item_iterator(this, query);
}
search_item *make_index_search_item(const char *filename, int fid)
{
char *index_filename = new char[strlen(filename) + sizeof(INDEX_SUFFIX)];
strcpy(index_filename, filename);
strcat(index_filename, INDEX_SUFFIX);
int fd = open(index_filename, O_RDONLY);
if (fd < 0)
return 0;
index_search_item *item = new index_search_item(index_filename, fid);
a_delete index_filename;
if (!item->load(fd)) {
close(fd);
delete item;
return 0;
}
else if (verify_flag && !item->verify()) {
delete item;
return 0;
}
else {
item->check_files();
return item;
}
}
index_search_item_iterator::index_search_item_iterator(index_search_item *ind,
const char *q)
: indx(ind), buf(0), buflen(0), temp_list(0), query(strsave(q)),
searcher(q, strlen(q), ind->ignore_fields, ind->header.truncate),
out_of_date_files_iter(0), next_out_of_date_file(0)
{
found_list = indx->search(q, strlen(q), &temp_list);
if (!found_list) {
found_list = &minus_one;
warning("all keys would have been discarded in constructing index `%1'",
indx->name);
}
}
index_search_item_iterator::~index_search_item_iterator()
{
a_delete temp_list;
a_delete buf;
a_delete query;
delete out_of_date_files_iter;
}
int index_search_item_iterator::next(const linear_searcher &,
const char **pp, int *lenp,
reference_id *ridp)
{
if (found_list) {
for (;;) {
int tagno = *found_list;
if (tagno == -1)
break;
found_list++;
if (get_tag(tagno, searcher, pp, lenp, ridp))
return 1;
}
found_list = 0;
next_out_of_date_file = indx->out_of_date_files;
}
while (next_out_of_date_file) {
if (out_of_date_files_iter == 0)
out_of_date_files_iter
= next_out_of_date_file->make_search_item_iterator(query);
if (out_of_date_files_iter->next(searcher, pp, lenp, ridp))
return 1;
delete out_of_date_files_iter;
out_of_date_files_iter = 0;
next_out_of_date_file = next_out_of_date_file->next;
}
return 0;
}
int index_search_item_iterator::get_tag(int tagno,
const linear_searcher &searcher,
const char **pp, int *lenp,
reference_id *ridp)
{
if (tagno < 0 || tagno >= indx->header.tags_size) {
error("bad tag number");
return 0;
}
tag *tp = indx->tags + tagno;
const char *filename = indx->munge_filename(indx->pool + tp->filename_index);
int fd = open(filename, O_RDONLY);
if (fd < 0) {
error("can't open `%1': %2", filename, strerror(errno));
return 0;
}
struct stat sb;
if (fstat(fd, &sb) < 0) {
error("can't fstat: %1", strerror(errno));
close(fd);
return 0;
}
time_t mtime = sb.st_mtime;
if (mtime > indx->mtime) {
indx->add_out_of_date_file(fd, filename,
indx->filename_id + tp->filename_index);
return 0;
}
int res = 0;
FILE *fp = fdopen(fd, "r");
if (!fp) {
error("fdopen failed");
close(fd);
return 0;
}
if (tp->start != 0 && fseek(fp, long(tp->start), 0) < 0)
error("can't seek on `%1': %2", filename, strerror(errno));
else {
int length = tp->length;
int err = 0;
if (length == 0) {
struct stat sb;
if (fstat(fileno(fp), &sb) < 0) {
error("can't stat `%1': %2", filename, strerror(errno));
err = 1;
}
else if (!S_ISREG(sb.st_mode)) {
error("`%1' is not a regular file", filename);
err = 1;
}
else
length = int(sb.st_size);
}
if (!err) {
if (length + 2 > buflen) {
a_delete buf;
buflen = length + 2;
buf = new char[buflen];
}
if (fread(buf + 1, 1, length, fp) != length)
error("fread on `%1' failed: %2", filename, strerror(errno));
else {
buf[0] = '\n';
buf[length + 1] = '\n';
res = searcher.search(buf + 1, buf + 2 + length, pp, lenp);
if (res && ridp)
*ridp = reference_id(indx->filename_id + tp->filename_index,
tp->start);
}
}
}
fclose(fp);
return res;
}
const char *index_search_item::munge_filename(const char *filename)
{
if (filename[0] == '/')
return filename;
const char *cwd = pool;
int need_slash = (cwd[0] != 0 && strchr(cwd, '\0')[-1] != '/');
int len = strlen(cwd) + strlen(filename) + need_slash + 1;
if (len > filename_buflen) {
a_delete filename_buffer;
filename_buflen = len;
filename_buffer = new char[len];
}
strcpy(filename_buffer, cwd);
if (need_slash)
strcat(filename_buffer, "/");
strcat(filename_buffer, filename);
return filename_buffer;
}
const int *index_search_item::search1(const char **pp, const char *end)
{
while (*pp < end && !csalnum(**pp))
*pp += 1;
if (*pp >= end)
return 0;
const char *start = *pp;
while (*pp < end && csalnum(**pp))
*pp += 1;
int len = *pp - start;
if (len < header.shortest)
return 0;
if (len > header.truncate)
len = header.truncate;
int is_number = 1;
for (int i = 0; i < len; i++)
if (csdigit(start[i]))
key_buffer[i] = start[i];
else {
key_buffer[i] = cmlower(start[i]);
is_number = 0;
}
if (is_number && !(len == 4 && start[0] == '1' && start[1] == '9'))
return 0;
unsigned hc = hash(key_buffer, len);
if (common_words_table) {
for (int h = hc % common_words_table_size;
common_words_table[h];
--h) {
if (strlen(common_words_table[h]) == len
&& memcmp(common_words_table[h], key_buffer, len) == 0)
return 0;
if (h == 0)
h = common_words_table_size;
}
}
int li = table[int(hc % header.table_size)];
return li < 0 ? &minus_one : lists + li;
}
static void merge(int *result, const int *s1, const int *s2)
{
for (; *s1 >= 0; s1++) {
while (*s2 >= 0 && *s2 < *s1)
s2++;
if (*s2 == *s1)
*result++ = *s2;
}
*result++ = -1;
}
const int *index_search_item::search(const char *ptr, int length,
int **temp_listp)
{
const char *end = ptr + length;
if (*temp_listp) {
a_delete *temp_listp;
*temp_listp = 0;
}
const int *first_list = 0;
while (ptr < end && (first_list = search1(&ptr, end)) == 0)
;
if (!first_list)
return 0;
if (*first_list < 0)
return first_list;
const int *second_list = 0;
while (ptr < end && (second_list = search1(&ptr, end)) == 0)
;
if (!second_list)
return first_list;
if (*second_list < 0)
return second_list;
const int *p;
for (p = first_list; *p >= 0; p++)
;
int len = p - first_list;
for (p = second_list; *p >= 0; p++)
;
if (p - second_list < len)
len = p - second_list;
int *matches = new int[len + 1];
merge(matches, first_list, second_list);
while (ptr < end) {
const int *list = search1(&ptr, end);
if (list != 0) {
if (*list < 0) {
a_delete matches;
return list;
}
merge(matches, matches, list);
if (*matches < 0) {
a_delete matches;
return &minus_one;
}
}
}
*temp_listp = matches;
return matches;
}
void index_search_item::read_common_words_file()
{
if (header.common <= 0)
return;
const char *common_words_file = munge_filename(strchr(pool, '\0') + 1);
errno = 0;
FILE *fp = fopen(common_words_file, "r");
if (!fp) {
error("can't open `%1': %2", common_words_file, strerror(errno));
return;
}
common_words_table_size = 2*header.common + 1;
while (!is_prime(common_words_table_size))
common_words_table_size++;
common_words_table = new char *[common_words_table_size];
for (int i = 0; i < common_words_table_size; i++)
common_words_table[i] = 0;
int count = 0;
int key_len = 0;
for (;;) {
int c = getc(fp);
while (c != EOF && !csalnum(c))
c = getc(fp);
if (c == EOF)
break;
do {
if (key_len < header.truncate)
key_buffer[key_len++] = cmlower(c);
c = getc(fp);
} while (c != EOF && csalnum(c));
if (key_len >= header.shortest) {
int h = hash(key_buffer, key_len) % common_words_table_size;
while (common_words_table[h]) {
if (h == 0)
h = common_words_table_size;
--h;
}
common_words_table[h] = new char[key_len + 1];
memcpy(common_words_table[h], key_buffer, key_len);
common_words_table[h][key_len] = '\0';
}
if (++count >= header.common)
break;
key_len = 0;
if (c == EOF)
break;
}
fclose(fp);
}
void index_search_item::add_out_of_date_file(int fd, const char *filename,
int fid)
{
search_item **pp;
for (pp = &out_of_date_files; *pp; pp = &(*pp)->next)
if ((*pp)->is_named(filename))
return;
*pp = make_linear_search_item(fd, filename, fid);
warning("`%1' modified since `%2' created", filename, name);
}
void index_search_item::check_files()
{
const char *pool_end = pool + header.strings_size;
for (const char *ptr = strchr(ignore_fields, '\0') + 1;
ptr < pool_end;
ptr = strchr(ptr, '\0') + 1) {
const char *path = munge_filename(ptr);
struct stat sb;
if (stat(path, &sb) < 0)
error("can't stat `%1': %2", path, strerror(errno));
else if (sb.st_mtime > mtime) {
int fd = open(path, O_RDONLY);
if (fd < 0)
error("can't open `%1': %2", path, strerror(errno));
else
add_out_of_date_file(fd, path, filename_id + (ptr - pool));
}
}
}